CoGraphNet for enhanced text classification using word-sentence heterogeneous graph representations and improved...

Area of Science:

• Natural Language Processing
• Machine Learning
• Graph Representation Learning

Background:

• Text Graph Representation Learning through Graph Neural Networks (TG-GNN) is effective but faces computational complexity and interpretability challenges.
• Existing methods may suffer from information loss in capturing multi-tiered contextual information.

Purpose of the Study:

• To propose CoGraphNet, a novel graph-based model for text classification that addresses computational complexity and interpretability issues.
• To enhance contextual comprehension and model clarity in text classification tasks.

Main Methods:

• Constructing separate heterogeneous graphs for words and sentences to capture multi-tiered contextual information.
• Incorporating positional bias weights to improve model interpretability and clarity.
• Utilizing novel graph structures and the SwiGLU activation function for enhanced contextual understanding.

Main Results:

• CoGraphNet demonstrates precise analysis by highlighting important words or sentences.
• Achieved enhanced contextual comprehension and accuracy compared to existing methods.
• Experimental validation on Ohsumed, MR, R52, and 20NG datasets confirms effectiveness.

Conclusions:

• CoGraphNet offers a superior approach to text classification by effectively managing computational complexity and improving interpretability.
• The model's novel graph structures and mechanisms provide enhanced contextual understanding, leading to superior performance in complex classification tasks.